\n\n
Endpoint Detection and Response for AI agents.
\nIntercept. Enforce. Observe. Every tool call. Every message. Every session.
\n\n\n\n ![\"npm\"](\"https://img.shields.io/npm/v/@cantinasecurity/clawsight?style=flat-square&logo=npm&label=npm\")
![\"License\"](\"https://img.shields.io/badge/license-MIT-blue?style=flat-square\"){kind=icon}
![\"OpenClaw](\"https://img.shields.io/badge/openclaw-plugin-orange?style=flat-square\"){kind=icon}
\n Getting Started ·\n Commands ·\n Architecture ·\n Default Rules ·\n Configuration\n
\n\n\n Architecture Docs ·\n Capabilities ·\n Full Config Reference ·\n Local Rules ·\n SIEM Rules ·\n API Reference\n
\n\n---\n\n## What is ClawSight?\n\nClawSight is an endpoint detection and response (EDR) plugin for OpenClaw AI agents. It hooks into the agent lifecycle — intercepting tool calls, outbound messages, and LLM interactions — and enforces security policies before actions execute.\n\nIt ships with 46 default rules covering reverse shells, credential theft, persistence mechanisms, exfiltration domains, and encoded payloads. You manage everything from Telegram (or Discord/Slack) using `/cs` commands.\n\n---\n\n## Getting Started\n\n### Option 1: Local policies only\n\nBest for: running an OpenClaw agent on your machine and managing security rules directly from Telegram. No external server needed. No data leaves your machine.\n\n**Step 1: Install the plugin**\n\n```bash\nnpx -y @cantinasecurity/clawsight install --mode local --link\n```\n\n**Step 2: Restart the gateway**\n\n```bash\nopenclaw gateway restart\n```\n\n**Step 3: Open Telegram and verify it's running**\n\nSend to your bot:\n```\n/cs status\n```\n\nYou should see:\n```\nClawSight status:\n Mode: local\n Rules loaded: 46\n ...\n```\n\n**That's it.** 46 rules are now active. Your agent can't run `curl`, access `.ssh/id_rsa`, create cron jobs, or connect to known exfiltration domains. You can add, remove, or modify any rule from Telegram.\n\nTry it:\n```\n/cs rules ← see all active rules\n/cs block domain evil.com ← block a domain\n/cs confirm command pip install ← require your approval before pip install\n/cs remove 5 ← remove a rule\n```\n\n---\n\n### Option 2: Connect to ClawSight SIEM\n\nBest for: streaming telemetry and enforcing remote policies via the ClawSight SIEM platform. The SIEM can run anywhere — a VPS, AWS, Vercel, or your local network.\n\n**Step 1: If your SIEM is on a remote server, set up an SSH tunnel**\n\n```bash\n# Forward local port 3000 to the SIEM running on your server\nssh -L 3000:127.0.0.1:3000 user@your-siem-server\n```\n\nSkip this if your SIEM has a public HTTPS URL (Vercel, AWS, etc.).\n\n**Step 2: Install with SIEM flags**\n\nVia SSH tunnel:\n```bash\nnpx -y @cantinasecurity/clawsight install \\\n --mode enforce \\\n --platform-url http://127.0.0.1:3000 \\\n --token YOUR_TOKEN \\\n --agent-name researcher-agent \\\n --link\n```\n\nOr directly to a public SIEM:\n```bash\nnpx -y @cantinasecurity/clawsight install \\\n --mode enforce \\\n --platform-url https://siem.example.com \\\n --token YOUR_TOKEN \\\n --agent-name researcher-agent \\\n --link\n```\n\n**Step 3: Restart and verify**\n\n```bash\nopenclaw gateway restart\n```\n\nTelemetry is now streaming to the SIEM — tool calls, messages, policy decisions, and agent inventory snapshots.\n\n\n## Managing Rules from Telegram\n\nAll rules are managed live from your chat channel. No config files to edit, no restarts needed.\n\n### See what's active\n\n```\n/cs status Show mode and rule counts\n/cs rules List all rules with IDs\n/cs directive preview See the full injected system prompt\n```\n\n### Block actions\n\n```\n/cs block command curl Block curl in shell commands\n/cs block command | bash Block piping to bash\n/cs block command base64 -d Block base64 decoding\n/cs block domain evil.com Block a domain + subdomains\n/cs block tool web_search Block a tool entirely\n/cs block command .aws/credentials Block access to AWS credentials\n```\n\n### Require your approval before an action runs\n\n```\n/cs confirm command pip install Require approval for pip install\n/cs confirm command npm install Require approval for npm install\n/cs confirm tool exec Require approval for all exec calls\n```\n\nWhen the agent tries a confirmed action, it's blocked and you get a pending ID:\n```\n/cs pending See pending approvals\n/cs approve a3f8 Approve (one-time)\n/cs approve-always a3f8 Approve + create permanent allow rule\n/cs deny a3f8 Deny\n```\n\n### Add instructions to the agent's prompt\n\n```\n/cs directive add Never share API keys in responses\n/cs directive add Always ask before deleting files\n/cs directives List all directives\n/cs directive remove 0 Remove by index\n```\n\n### Remove or reset rules\n\n```\n/cs remove 5 Remove rule #5\n/cs remove 5 6 7 Remove multiple\n/cs reset confirm Reset everything to defaults\n```\n\n---\n\n## Architecture\n\n```mermaid\ngraph LR\n User[\"User